Towards Superintelligence?
XVI Latin American Congress of Statistical Societies (CLATSE) · Quito, Ecuador · October 2025
In October 2025, Quito became the epicenter of Latin American statistics. The XVI Latin American Congress of Statistical Societies (CLATSE) brought together academics, researchers, and professionals from across the region for five days to debate the present and future of data analysis. In that setting, the talk "Towards Superintelligence?" offered a candid look at artificial intelligence: what it can really do, where it fails, what risks it brings, and where it is headed.
The session opened with a simple but uncomfortable question: do we truly understand what AI does, or are we merely watching it with wonder? From there, the talk explored the actual workings of large language models (LLMs), dismantling the notion that "predicting text" is the same as "understanding." Confusing the two has real consequences.
The risks are not science fiction. Electoral manipulation, algorithmic bias amplified at scale, deepfakes, and the use of social media as a disinformation weapon — with Myanmar as a stark example of real human consequences — were all part of the journey. But the talk did not dwell in pessimism: it also explored AI's genuine successes, from chess and Go to unexpected forms of creativity, alongside the problem of reward hacking (when a system achieves its goal… but not quite as intended).
The second half looked forward. What happens when automation displaces jobs faster than institutions can absorb? How do you govern a system smarter than those who created it? The Gorilla dilemma and the Midas dilemma illustrated the central challenge of superintelligence: building systems whose goals are, genuinely, to help us — not merely to execute instructions to the letter.
The closing was a call to action: aligned and beneficial AI is not an automatic outcome; it is a design choice that demands urgency, rigor, and human participation.
Workshop: Practical AI for Curious Humans
Participation at the XVI CLATSE was not limited to the keynote. As part of the congress, the workshop "Practical AI for Curious Humans" was also delivered — a space to explore the tools and concepts from the talk in a direct, hands-on way.
Interactive Presentation
The presentation was built with Streamlit and Reveal.js (Python 3.12) as an interactive experience — not a static slide deck. It is available online for anyone who wants to explore it at their own pace.